Support Antigravity SDK as a first-class agent implementation framework

[yu-iskw]

What is your feature suggestion?

lease add first-class support for agents implemented with the Google Antigravity SDK for Python (google-antigravity) in agents-cli.

Today, agents-cli already works well as a lifecycle toolchain for Google Cloud agent development, and it already documents support for Antigravity as a coding assistant that can use the CLI. This feature request is about a different layer: using Antigravity SDK as the framework/runtime for the agent being created, run, evaluated, deployed, and published by agents-cli.

A useful first version could be an Antigravity SDK project template:

agents-cli create my-antigravity-agent --agent antigravity-sdk-python

The generated project should include a minimal runnable Antigravity SDK agent and the surrounding agents-cli project structure needed for local development, deployment, evaluation, and documentation.

Desired capabilities, ideally delivered incrementally:

1. Built-in Antigravity SDK template

   • Add an antigravity-sdk-python template alongside the existing agent templates.
   • Generate a minimal but production-shaped Python project using google-antigravity.
   • Include a runnable Agent / LocalAgentConfig example.
   • Include examples for a custom Python tool and an optional MCP server configuration.

2. Manifest support

   • Allow agents-cli-manifest.yaml to identify the agent framework as Antigravity SDK.
   • Capture the Antigravity agent entrypoint, runtime/deployment target, environment variables, and supported lifecycle commands.

3. Local run and playground support

   • Support running the generated Antigravity SDK agent locally through agents-cli run or document the supported local runner contract.
   • Support local development with either a Gemini API key or Vertex / Gemini Enterprise Agent Platform configuration where applicable.

4. Evaluation support

   • Support agents-cli eval for Antigravity SDK agents directly, or document an adapter interface that normalizes Antigravity SDK responses into the existing agents-cli evaluation format.
   • If Antigravity SDK exposes streaming steps, tool calls, or traces, document how much of that information can be collected for evaluation.

5. Deployment and publishing guidance

   • Document and test at least one supported deployment target, such as Cloud Run, Agent Runtime, or GKE.
   • Clarify whether Antigravity SDK agents can be published to Gemini Enterprise Agent Platform through agents-cli publish gemini-enterprise.
   • Provide safe defaults for service accounts, environment variables, secrets, ADC, and MCP configuration.

6. Safety and governance examples

   • Include examples for Antigravity SDK hooks or policies, such as read-only tools, deny-by-default tool policy, or explicit approval before risky tool execution.
   • Document recommended defaults for enterprise projects that use SaaS, public-cloud, or MCP tools.

What will this enable you to do?

This would allow teams to use agents-cli as the standard Google Cloud lifecycle toolchain while choosing Antigravity SDK as the agent implementation framework.

Example workflows this would enable:

agents-cli create incident-triage-agent --agent antigravity --deployment-target cloud_run
agents-cli run "Summarize this incident and propose next actions"
agents-cli eval run
agents-cli deploy
agents-cli publish gemini-enterprise

More specifically, this would enable developers to:

1. Build Antigravity SDK agents with production scaffolding

   • Start from a supported project layout instead of hand-rolling the integration.
   • Use the same agents-cli lifecycle conventions as other Google Cloud agent projects.

2. Use Antigravity SDK capabilities in enterprise agents

   • Implement agents with Antigravity SDK primitives.
   • Use stateful conversations and streaming responses.
   • Register custom Python tools.
   • Integrate MCP servers as tool providers.
   • Apply hooks and policies for safer tool execution.
   • Configure local Gemini API key development and Google Cloud / Vertex / Gemini Enterprise modes where supported.

3. Reduce framework fragmentation

   • Avoid forcing teams to choose between ADK-oriented agents-cli lifecycle support and Antigravity SDK-specific agent capabilities.
   • Make it clear when to choose ADK, Antigravity SDK, or another framework for a given agent project.

4. Improve enterprise governance

   • Give platform teams a consistent way to scaffold, evaluate, deploy, and review Antigravity SDK agents.
   • Encourage safer defaults around credentials, service accounts, MCP tools, and production deployment.

5. Support phased adoption

   • Teams could begin with a supported Antigravity SDK template and later adopt deeper agents-cli integration as run/eval/deploy/publish support matures.

Additional context

I think this is worthwhile because agents-cli is positioned as the lifecycle CLI for building, evaluating, deploying, publishing, governing, and optimizing agents on Google Cloud, while Antigravity SDK is a Google Python SDK for building agents powered by Antigravity and Gemini. Developers may reasonably expect these two Google agent-development surfaces to work together not only at the coding-assistant layer, but also at the agent-implementation-framework layer.

Current context that motivated this request:

• agents-cli already documents support for Antigravity as a coding assistant that can use the CLI.
• The currently documented built-in templates appear to be ADK-oriented, such as adk, adk_a2a, and agentic_rag.
• agents-cli create --agent already supports template identifiers, local paths, ADK sample shortcuts, and remote Git URLs, which suggests a low-risk incremental path: start with a documented remote Antigravity SDK starter template, then promote it to a first-class built-in template if it proves useful.
• Antigravity SDK has its own agent API/runtime surface, including agent configuration, conversations, streaming, custom tools, MCP integration, hooks/policies, triggers, and Google Cloud / Vertex / Gemini Enterprise configuration.